aboutsummaryrefslogtreecommitdiff
path: root/src/cpu/kernels/CpuConcatenateWidthKernel.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/cpu/kernels/CpuConcatenateWidthKernel.cpp')
-rw-r--r--src/cpu/kernels/CpuConcatenateWidthKernel.cpp118
1 files changed, 65 insertions, 53 deletions
diff --git a/src/cpu/kernels/CpuConcatenateWidthKernel.cpp b/src/cpu/kernels/CpuConcatenateWidthKernel.cpp
index f00b37a01b..f6100cccca 100644
--- a/src/cpu/kernels/CpuConcatenateWidthKernel.cpp
+++ b/src/cpu/kernels/CpuConcatenateWidthKernel.cpp
@@ -24,12 +24,12 @@
#include "src/cpu/kernels/CpuConcatenateWidthKernel.h"
#include "arm_compute/core/Error.h"
-#include "arm_compute/core/Steps.h"
#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/Steps.h"
#include "arm_compute/core/Validate.h"
-#include "src/core/NEON/NEAsymm.h"
-#include "src/core/helpers/WindowHelpers.h"
+#include "src/core/helpers/WindowHelpers.h"
+#include "src/core/NEON/NEAsymm.h"
namespace arm_compute
{
@@ -47,7 +47,7 @@ Status validate_arguments(const ITensorInfo *src, unsigned int width_offset, con
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
ARM_COMPUTE_RETURN_ERROR_ON(src->dimension(0) + width_offset > dst->dimension(0));
- for(size_t i = 1; i < Coordinates::num_max_dimensions; ++i)
+ for (size_t i = 1; i < Coordinates::num_max_dimensions; ++i)
{
ARM_COMPUTE_RETURN_ERROR_ON(src->dimension(i) != dst->dimension(i));
}
@@ -86,13 +86,14 @@ void CpuConcatenateWidthKernel::run_op(ITensorPack &tensors, const Window &windo
auto dst = tensors.get_tensor(TensorType::ACL_DST);
// Offset output pointer to the correct position
- uint8_t *dst_ptr = dst->buffer() + dst->info()->offset_first_element_in_bytes() + _width_offset * dst->info()->strides_in_bytes()[0];
+ uint8_t *dst_ptr = dst->buffer() + dst->info()->offset_first_element_in_bytes() +
+ _width_offset * dst->info()->strides_in_bytes()[0];
const auto window_start_x = static_cast<int>(window.x().start());
const auto window_end_x = static_cast<int>(window.x().end()) * static_cast<int>(dst->info()->element_size());
constexpr int window_step_x = 16;
- Window win{ window };
+ Window win{window};
win.set(Window::DimX, Window::Dimension(0, 1, 1));
// Create iterators
@@ -101,62 +102,73 @@ void CpuConcatenateWidthKernel::run_op(ITensorPack &tensors, const Window &windo
const DataType dt = src->info()->data_type();
const UniformQuantizationInfo &src_qinfo = src->info()->quantization_info().uniform();
const UniformQuantizationInfo &dst_qinfo = dst->info()->quantization_info().uniform();
- if(dt == DataType::QASYMM8 && src_qinfo != dst_qinfo)
+ if (dt == DataType::QASYMM8 && src_qinfo != dst_qinfo)
{
- execute_window_loop(win, [&](const Coordinates &)
- {
- int x = window_start_x;
- for(; x <= (window_end_x - window_step_x); x += window_step_x)
+ execute_window_loop(
+ win,
+ [&](const Coordinates &)
{
- vst1q_u8(dst_ptr + dst_it.offset() + x, vquantize(vdequantize(vld1q_u8(src_it.ptr() + x), src_qinfo), dst_qinfo));
- }
-
- // Compute left-over elements
- for(; x < window_end_x; ++x)
- {
- *(dst_ptr + dst_it.offset() + x) = quantize_qasymm8(dequantize_qasymm8(*(src_it.ptr() + x), src_qinfo), dst_qinfo);
- }
- },
- src_it, dst_it);
+ int x = window_start_x;
+ for (; x <= (window_end_x - window_step_x); x += window_step_x)
+ {
+ vst1q_u8(dst_ptr + dst_it.offset() + x,
+ vquantize(vdequantize(vld1q_u8(src_it.ptr() + x), src_qinfo), dst_qinfo));
+ }
+
+ // Compute left-over elements
+ for (; x < window_end_x; ++x)
+ {
+ *(dst_ptr + dst_it.offset() + x) =
+ quantize_qasymm8(dequantize_qasymm8(*(src_it.ptr() + x), src_qinfo), dst_qinfo);
+ }
+ },
+ src_it, dst_it);
}
- else if(dt == DataType::QASYMM8_SIGNED && src_qinfo != dst_qinfo)
+ else if (dt == DataType::QASYMM8_SIGNED && src_qinfo != dst_qinfo)
{
- execute_window_loop(win, [&](const Coordinates &)
- {
- int x = window_start_x;
- for(; x <= (window_end_x - window_step_x); x += window_step_x)
- {
- vst1q_s8(reinterpret_cast<int8_t *>(dst_ptr + dst_it.offset() + x),
- vquantize_signed(vdequantize(vld1q_s8(reinterpret_cast<int8_t *>(src_it.ptr() + x)), src_qinfo), dst_qinfo));
- }
-
- // Compute left-over elements
- for(; x < window_end_x; ++x)
+ execute_window_loop(
+ win,
+ [&](const Coordinates &)
{
- *(dst_ptr + dst_it.offset() + x) = quantize_qasymm8_signed(dequantize_qasymm8_signed(*(src_it.ptr() + x), src_qinfo), dst_qinfo);
- }
- },
- src_it, dst_it);
+ int x = window_start_x;
+ for (; x <= (window_end_x - window_step_x); x += window_step_x)
+ {
+ vst1q_s8(
+ reinterpret_cast<int8_t *>(dst_ptr + dst_it.offset() + x),
+ vquantize_signed(vdequantize(vld1q_s8(reinterpret_cast<int8_t *>(src_it.ptr() + x)), src_qinfo),
+ dst_qinfo));
+ }
+
+ // Compute left-over elements
+ for (; x < window_end_x; ++x)
+ {
+ *(dst_ptr + dst_it.offset() + x) =
+ quantize_qasymm8_signed(dequantize_qasymm8_signed(*(src_it.ptr() + x), src_qinfo), dst_qinfo);
+ }
+ },
+ src_it, dst_it);
}
else
{
- execute_window_loop(win, [&](const Coordinates &)
- {
- const auto in_ptr = src_it.ptr();
- const auto out_ptr = dst_ptr + dst_it.offset();
- int x = window_start_x;
- for(; x <= (window_end_x - window_step_x); x += window_step_x)
- {
- wrapper::vstore(out_ptr + x, wrapper::vloadq(in_ptr + x));
- }
-
- // Compute left-over elements
- for(; x < window_end_x; ++x)
+ execute_window_loop(
+ win,
+ [&](const Coordinates &)
{
- *(out_ptr + x) = *(in_ptr + x);
- }
- },
- src_it, dst_it);
+ const auto in_ptr = src_it.ptr();
+ const auto out_ptr = dst_ptr + dst_it.offset();
+ int x = window_start_x;
+ for (; x <= (window_end_x - window_step_x); x += window_step_x)
+ {
+ wrapper::vstore(out_ptr + x, wrapper::vloadq(in_ptr + x));
+ }
+
+ // Compute left-over elements
+ for (; x < window_end_x; ++x)
+ {
+ *(out_ptr + x) = *(in_ptr + x);
+ }
+ },
+ src_it, dst_it);
}
}
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-19 21:48:16 +0000